Regional assessment of boreal forest productivity using an ecological process model and remote sensing parameter maps

Transcription

1 Tree Physiology 20, Heron Publishing Victoria, Canada Regional assessment of boreal forest productivity using an ecological process model and remote sensing parameter maps J. S. KIMBALL, 1,3 A. R. KEYSER, 1 S. W. RUNNING 1 and S. S. SAATCHI 2 1 Numerical Terradynamic Simulation Group, School of Forestry, University of Montana, Missoula, MT 59812, USA 2 NASA Jet Propulsion Laboratory, Pasadena, CA 91109, USA 3 Current address: Flathead Lake Biological Station, University of Montana, 311 Bio Station Lane, Polson, MT , USA Received June 3, 1999 Summary An ecological process model (BIOME-BGC) was used to assess boreal forest regional net primary production (NPP) and response to short-term, year-to-year weather fluctuations based on spatially explicit, land cover and biomass maps derived by radar remote sensing, as well as soil, terrain and daily weather information. Simulations were conducted at a 30-m spatial resolution, over a 1205 km 2 portion of the BOREAS Southern Study Area of central Saskatchewan, Canada, over a 3-year period ( ). Simulations of NPP for the study region were spatially and temporally complex, averaging 2.2 (± 0.6), 1.8 (± 0.5) and 1.7 (± 0.5) Mg C ha 1 year 1 for 1994, 1995 and 1996, respectively. Spatial variability of NPP was strongly controlled by the amount of aboveground biomass, particularly photosynthetic leaf area, whereas biophysical differences between broadleaf deciduous and evergreen coniferous vegetation were of secondary importance. Simulations of NPP were strongly sensitive to year-to-year variations in seasonal weather patterns, which influenced the timing of spring thaw and deciduous bud-burst. Reductions in annual NPP of approximately 17 and 22% for 1995 and 1996, respectively, were attributed to 3- and 5-week delays in spring thaw relative to Boreal forest stands with greater proportions of deciduous vegetation were more sensitive to the timing of spring thaw than evergreen coniferous stands. Similar relationships were found by comparing simulated snow depth records with 10-year records of aboveground NPP measurements obtained from biomass harvest plots within the BOREAS region. These results highlight the importance of sub-grid scale land cover complexity in controlling boreal forest regional productivity, the dynamic response of the biome to short-term interannual climate variations, and the potential implications of climate change and other large-scale disturbances. Keywords: BIOME-BGC, BOREAS, carbon cycle, ecosystem modeling, net primary production, NPP, radar, SAR. Introduction The boreal forest biome occupies between 12 and 20 million km 2 and plays an important role in the global carbon cycle (Whittaker and Likens 1975, Schlesinger 1997). Seasonal and interannual variations in temperate and high latitude (30 60 N) terrestrial net primary production (NPP) are thought to be major factors regulating global atmospheric CO 2 concentrations (Ciais et al. 1995, Kindermann et al. 1996). Recent evidence indicates a trend toward increased amplitude of the seasonal atmospheric CO 2 cycle and enhanced photosynthetic activity of terrestrial vegetation, linked to higher spring temperatures, earlier seasonal snowmelt and a lengthening of the active growing season for temperate and boreal regions (Keeling et al. 1996, Myneni et al. 1997). The boreal forest represents a complex land cover mosaic where vegetation morphology, condition and distribution are strongly regulated by environmental factors such as water availability, growing season length, disturbance and soil nutrients (Bonan and Shugart 1989). A major research effort has been directed toward improving our understanding of the linkages among climatological, physical, carbon uptake and respiration processes within the boreal biome (Sellers et al. 1997). Much of this research has been directed at detailed studies of carbon dynamics for individual stands (e.g., Black et al. 1996, Frolking et al. 1996, Baldocchi et al. 1997, Jarvis et al. 1997, Goulden et al. 1998). Extrapolation of these results to characterize large regions is difficult, however, because the dynamics of carbon sequestration processes are spatially and temporally complex. As a result, the integrated effects of these processes are still poorly understood at regional scales. A fundamental strategy of the Boreal Ecosystem Atmosphere Study (BOR- EAS) and other intensive large-scale field experiments has been to extend plot scale relationships to more regional scales, using process models coupled with remote sensing and other integrative modeling techniques. In this investigation, we assessed the relative magnitude and spatial complexity of annual NPP across a boreal landscape, and the sensitivity of the system to short-term, inter-annual weather variations. For this purpose, we used an ecosystem process model coupled with land cover and biomass maps derived by radar remote sensing, terrain data and daily weather information. Regional simulations were conducted at a 30-m

2 762 KIMBALL, KEYSER, RUNNING AND SAATCHI spatial resolution, over a 1205 km 2 portion of the BOREAS Southern Study Area (SSA) of central Saskatchewan, Canada, over a 3-year period ( ). We also compared 10-year ( ) annual aboveground NPP measurements with simulated spring snow depletion dates for different forest types within the BOREAS study areas, to investigate relationships between NPP and spring thaw timing, and interpret seasonal and inter-annual behavior in regional simulations. Materials and methods Study area The BOREAS Southern Modeling Sub-Area (SMSA) covers an area of approximately km within the BOREAS SSA (~53 55 N, W) and is fairly typical of the southern boreal forest margin (BOREAS Science Team 1995, Sellers et al. 1997). Topographic variability is low, with elevations ranging from 440 to 660 m and slopes generally less than 5%. Vegetation of the region includes both deciduous and coniferous life forms. Dry, sandy upland sites support jack pine (Pinus banksiana Lamb.) stands, whereas aspen (Populus tremuloides Michx.), balsam poplar (Populus balsamifera L.) and white spruce (Picea glauca (Moench.) Voss) are found on welldrained, glacial deposits. In wet, poorly drained areas, black spruce (Picea mariana (Mill.) BSP) and tamarack (Larix laricina (Du Roi) C. Koch) are common. Bogs and fens are also common in poorly drained areas, and are primarily composed of sedges (Carex spp.) interspersed with black spruce, tamarack and bog birch (Betula pumila L.). Logging and fire-related disturbances also play a major role in shaping vegetation patterns in the area. Localized logging for paper pulp and fence posts is common along roadside areas within the study area, whereas the northeast portion of the SMSA encompasses part of an extensive burn that occurred in 1977 and This area is predominantly covered with small (< 5 m) jack pine regrowth. There was another fire in 1995 and more logging the following year. However, changes in land cover characteristics after 1994 were not considered in this investigation. Ecosystem model description BIOME-BGC (BioGeoChemistry) is a process-level, ecosystem model that simulates biogeochemical and hydrologic variables within multiple biomes. Model logic is based on the assumption that differences in process rates among biomes are primarily a function of climate and general life-form characteristics (Running and Hunt 1993). The model represents a compromise between: (1) the desire to represent detailed surface structure and biophysical interactions readily observed at the plot scale, and (2) the limitations of surface biophysical, meteorological and validation information at regional scales. The model employs several simplifying strategies regarding stand and meteorological conditions to facilitate application at regional scales. Each surface unit is represented by single, homogeneous canopy, snow and soil layers, where understory processes are not distinguished from the aggregate. Meteorological characteristics are defined from daily minimum and maximum air temperatures, precipitation and solar irradiance. These data are used in conjunction with general stand and soil information to predict net photosynthesis, respiration, evapotranspiration, snow cover and soil water conditions on a daily basis. BIOME-BGC logic, input requirements and applications for various environments, biomes and spatial scales are well documented elsewhere (e.g., Hunt and Running 1992, Running and Hunt 1993, Hunt et al. 1996). Documentation of model structure and input requirements unique to the BOREAS environment are also provided by Kimball et al. (1997a, 1997b). A summary of model structure relating to the characterization of spatially distributed carbon fluxes within the SMSA is provided here. Net primary production represents the net accumulation of carbon by the stand, and is determined as the daily difference between gross photosynthesis and respiration from autotrophic maintenance (R m ) and growth (R g ) processes. Photosynthesis is calculated by a modified form of the Farquhar biochemical model (Farquhar and von Caemmerer 1982, Kimball et al. 1997a). Photosynthetic response is regulated by canopy conductance to CO 2, leaf maintenance respiration and daily meteorological conditions including air pressure, air temperature and solar irradiance. Canopy CO 2 conductance is calculated as a proportion (62.5%) of the canopy conductance to water vapor (g c ). The maximum canopy water vapor conductance (g c,max ) defines the upper boundary of the photosynthetic rate, and is determined by leaf area index (LAI), prescribed leaf-scale boundary layer (g bl ) and stomatal minimum (g st,min ) and maximum (g st,max ) conductances to water vapor; g c is reduced in a nonlinear fashion when air temperature (T a ), vapor pressure deficit (VPD), solar irradiance or soil water potential (PSI) deviate from prescribed optimal conditions (Running and Hunt 1993, Kimball et al. 1997a, 1997b). The R m term represents the total loss of carbon from the system as a result of day and night foliar (R dl + R nl ), sapwood (R sw ), coarse root (R cr ) and fine root (R fr ) respiration components. The R m term is calculated from mean daily air temperatures and prescribed foliar, root (coarse + fine) and stem carbon pools based on an exponential relationship between respiration and temperature (Kimball et al. 1997a). The magnitude of the respiration response to temperature is governed by a prescribed rate defined at a reference temperature (i.e., 20 C) and a proportional rate change for a 10 C change in temperature (Q 10 ). Daily growth respiration was not determined explicitly by the model, but was computed as a proportion (32%) of the daily difference between gross photosynthesis and R m (Penning de Vries et al. 1974, Lavigne and Ryan 1997). Evapotranspiration is computed as the daily sum of transpiration and evaporation from surface, snow and canopy components. Both transpiration and evaporation components are estimated from daily air temperature, humidity and solar irradiance information, using a modified Penman-Monteith approach (Running and Hunt 1993, Kimball et al. 1997b). Maximum transpiration rates are regulated by g c. Surface evaporation is controlled by a surface conductance term that deviates TREE PHYSIOLOGY VOLUME 20, 2000

3 REGIONAL ASSESSMENT OF BOREAL NPP 763 from an optimal rate by an inverse exponential decay function based on the number of days since a rainfall event (Kimball et al. 1997b). Daily precipitation is categorized as rain or snow based on an air temperature threshold of 0 C. Rainfall is intercepted by the canopy using a prescribed interception coefficient based on LAI. Intercepted precipitation is then evaporated from the canopy using a Penman combination method and a prescribed boundary layer conductance (Running and Coughlan 1988). All remaining rainfall is routed directly to the surface. Snowfall is not intercepted by the vegetation canopy and is passed directly to the surface. Snowfall is stored as depth of water equivalent and no attempt is made to account for snowpack depth or density. If the mean daily air temperature is less than 0 C, snowpack sublimation is derived from the estimated daily net solar radiation (R n,s ) at the surface. If the air temperature is greater than 0 C, snowmelt is calculated by a degree-day method and R n,s. BIOME-BGC uses daily maximum and minimum air temperatures, humidity, incident solar radiation and precipitation as the fundamental environmental drivers of ecological processes. Incident short-wave radiation (direct + diffuse) is simulated using MT-CLIM logic as described by Running et al. (1987). Estimated total daily incident radiation is attenuated through the vegetation canopy using Beer s formulation and a prescribed extinction coefficient modulated by LAI. The R n,s term is estimated with prescribed albedos for snow (0.8) and vegetated surfaces. Maximum and minimum daily air temperatures are used to estimate mean daily air temperatures. Minimum daily air temperature is assumed equal to the dew point and is used with mean daily air temperature to estimate the mean daily VPD. These results are used with estimated day length and canopy biophysical information to compute hydrologic and carbon budgets on a daily basis. Since its inception as a point-scale model, BIOME-BGC has evolved to simulate regional scale processes by incorporating spatially distributed daily meteorological fields derived from a microclimate simulator, and remote sensing derived surface parameter maps to define important landscape characteristics. The model employs a biome-level stratification of land cover conditions to minimize spatial variability in conversion efficiencies and potential environmental controls, and is now capable of simulating landscape ecosystem processes (Running et al. 1989, Running and Hunt 1993, Hunt et al. 1996). BIOME-BGC stand-level simulations were previously conducted for five BOREAS black spruce, jack pine and aspen tower eddy flux sites for 1994 (Kimball et al. 1997a,1997b). Model simulations explained 98, 62 and 66% of respective variations in daily soil water, evapotranspiration and net daily CO 2 exchange measurements, and were found to be generally consistent with measured results, given the ranges of uncertainty and intra-stand variability in model input variables and measured processes. Both model results and field measurements have shown that photosynthesis, respiration, NPP and net CO 2 exchange are strongly sensitive to environmental controls such as VPD, soil and air temperatures (e.g., Fan et al. 1995, Dang et al. 1997a, Hogg and Hurdle 1997, Goulden et al. 1998). These studies also show that the relative magnitudes of ecosystem processes and sensitivities to environmental controls differ markedly among major boreal vegetation life forms (e.g., broadleaf (deciduous aspen) and evergreen (coniferous black spruce and jack pine)) and age classes. Land cover conditions and environmental controls within boreal regions are spatially and temporally complex (Bonan and Shugart 1989, Saatchi and Rignot 1997). All of these factors inhibit the inference of regional ecosystem processes from detailed plot-level measurements covering limited time periods. Below we discuss an approach for regional estimation of NPP using BIOME-BGC, coupled with spatially explicit daily meteorological, terrain and land cover information. Model inputs and initialization For landscape simulations, BIOME-BGC uses a spatial database composed of elevation, soil, vegetation and daily meteorological characteristics registered to a common projection format, as well as an array of critical physiological constants that define the environmental response curves of individual biome types within the spatial domain (Table 1). These physiological constants were obtained from BOREAS field measurements when possible. When these data were unavailable, values were selected from the literature for representative cover types under similar environmental conditions. Remote sensing-derived crown and stem biomass and land cover maps were used to define surface structure for the model simulations. These maps were derived from classification of airborne synthetic aperture radar (AIRSAR) remote sensing measurements of the SMSA acquired on July 21, AIRSAR is a three-frequency SAR operating at P-band (68 cm wavelength), L-band (24 cm) and C-band (5.6 cm) frequencies with HH, HV and VV polarization. The SAR data are particularly useful for extracting biomass and land cover information in boreal regions because of the sensitivity of radar to vegetation structural characteristics and insensitivity to low solar irradiance, smoke and cloud cover, which are problematic for optical remote sensing methods (Waring et al. 1995). Approximately 15 synoptic (50-km swath) AIRSAR images were acquired from a DC-8 aircraft and processed into a 30-m spatial resolution mosaic over the SMSA with less than a 0.1-dB absolute calibration error (Saatchi and Rignot 1997). The spatial extent of the AIRSAR maps defined the study area for the regional simulations, which were limited to approximately 1205 km 2 within the SMSA (Figure 1). Detailed information regarding AIRSAR acquisition, processing and classification of land cover and biomass within the SMSA are provided elsewhere (e.g., BOREAS Science Team 1995, Saatchi and Rignot 1997, Saatchi and Moghaddam 2000a, 2000b). A discussion of methods for incorporating this information within the framework of a regional ecosystem process model for estimating NPP is presented below. The AIRSAR land cover classification map defined the number of individual biome types represented in the ecosys- TREE PHYSIOLOGY ON-LINE at

4 764 KIMBALL, KEYSER, RUNNING AND SAATCHI Table 1. Summary of BIOME-BGC constants for coniferous and deciduous vegetation types within the BOREAS SMSA; constants were estimated from both BOREAS field measurements and literature sources for representative cover types. Parameter Coniferous Deciduous Reference Leaf N (%) , 13, 14 Leaf N in Rubisco (%) , 2 Max. g s (mm s 1 ) , 4 g bl (mm s 1 ) Optimal air temperature for g s ( C) , 11 Ψ at stomatal closure ( Mpa) VPD at stomatal closure (kpa) , 4 Albedo (no snow conditions, %) Leaf, fine root R m proportions at 20 C (1/day) , 9, 10 Stem, coarse root R m proportions at 20 C (1/day) , 9, 10 Canopy extinction of PAR , 12 Q 10 for R m Foliar/crown biomass Ratio of total to projected LAI , 7 SLA (1-sided, m 2 kg 1 ) Coarse root C/live stem C , 20 Fine root C/foliar C , 16, 17 Soil water holding capacity (%) , 19 Effective soil depth (m) , 19 References: 1, Field and Mooney 1986; 2, Fan et al. 1995; 3, Dang et al. 1997a; 4, Hogg and Hurdle 1997; 5, Waring and Running 1998; 6, Baldocchi et al. 1997; 7, Sprugel et al. 1995; 8, Betts and Ball 1997; 9, Vowinckel et al. 1975; 10, Johnson-Flanagan and Owens 1986; 11, Lavigne and Ryan 1997; 12, Dang et al. 1997b; 13, Sullivan et al. 1997; 14, Aerts et al. 1992; 15, Gower et al. 1997; 16, Steele et al. 1997; 17, Mitsch and Gosselink 1993; 18, Acton et al. 1991; 19, Cuenca et al. 1997; 20, Vogt tem model, whereas biomass maps were used to determine LAI and foliar and stem carbon pools within each grid cell. Seven land cover classes were distinguished within the SMSA, representing dry conifer (DC), wet conifer (WC), open water (OW), disturbed (DSTRB), deciduous (DEC), mixed deciduous conifer (MX) and wetland (WD) areas. The DC areas were mainly composed of jack pine stands, whereas WC areas consisted mainly of black spruce stands. The WD areas were composed of a mixture of black spruce dominated forest, bog and fen sites, whereas DSTRB sites represented a mixture of recently logged or burned areas, roads and other sparsely vegetated, non-water surfaces. The DEC areas generally had more than 80% deciduous cover dominated by aspen stands and grassland, whereas MX areas represented a mixture of mainly jack pine and aspen forest with no clear dominance of either. Open water and non-vegetated areas were not represented in the model and were masked from further analysis. Deciduous and coniferous canopies within MX cells were simulated sep- Figure 1. Map of the BOREAS Southern Study Area (SSA). The study region (i.e., modeling grid) for this investigation represented an area of approximately 1205 km 2 within the BOREAS Southern Modeling Sub- Area (SMSA) and was defined by the spatial extent of available AIRSAR-derived biomass and land cover information. TREE PHYSIOLOGY VOLUME 20, 2000

5 REGIONAL ASSESSMENT OF BOREAL NPP 765 arately, because of differences in biophysical characteristics and physiological responses to environmental controls. The DC and WC classes were assumed to be composed entirely of coniferous vegetation, whereas DEC areas were assumed to be 100% deciduous. The DSTRB, MX and WD classes were represented as a mixture of 50% deciduous and 50% coniferous life forms. Ideally, information on deciduous and coniferous proportional cover characteristics could be used to simulate the relative contributions of each to the total water and carbon flux within each cell. Unfortunately, that information was not available for this investigation, making simplifying assumptions necessary. Living stem carbon mass was derived from the AIRSAR stem biomass map and estimates of the relative proportions of carbon in stem biomass and living cells in sapwood tissue. This information was obtained from BOREAS biomass harvest plot data and other information reported in the literature for representative vegetation types (Gower et al. 1997, Waring and Running 1998). Living coarse root carbon mass was estimated as a proportion (0 25%) of live stem carbon based on allometric relationships for representative cover types (Grier et al. 1981, Vogt 1991, Steele et al. 1997). Living, fine-root carbon mass was estimated as 1.5 to 3.0 times foliar carbon estimates based on BOREAS biomass measurements and information reported in the literature for nutrient-limited arctic, boreal and cold temperate environments (Bigger and Oechel 1982, Mitsch and Gosselink 1993, Schimel et al. 1996, Gower et al. 1997, Steele et al. 1997). Foliar carbon mass was derived from AIRSAR crown biomass (i.e., leaves + branches) maps and estimated proportions of foliar to crown biomass obtained from BOREAS biomass harvest plots (Gower et al. 1997). Leaf area index (LAI) was derived from foliar carbon maps and specific leaf area (SLA) values were derived from BOREAS canopy biophysical measurements (Dang et al. 1997b). Values of LAI for coniferous vegetation were held constant over each year. For deciduous vegetation, LAI was regulated between a prescribed seasonal minimum (i.e., 0.0) and the remote sensing defined seasonal maximum using a phenology model based on daily meteorological predictions of satellite-observed dates of greenness onset and offset (White et al. 1997). The model predicts the onset of greenness from a combined thermal and radiation summation, whereas offset is determined from a thermally adjusted photoperiod trigger. The foliar carbon pool was increased on a daily basis using a stepped, 45-day linear ramping function between the onset date and the AIRSAR-defined LAI, whereas foliage drop occurred at the offset date. Foliar leaf nitrogen concentrations (N leaf ) strongly influence the photosynthetic capacity of the system and are directly related to the amount of radiation absorbed by the canopy (Schimel et al. 1991, Pierce et al. 1994). The N leaf term is used with canopy absorbed solar radiation, leaf maintenance respiration and CO 2 conductance information to compute daily carbon assimilation in BIOME-BGC. Because canopy absorption is also related to LAI, N leaf was estimated as a proportion of foliar biomass for each land cover class. These fractions were derived from site measurements within BOREAS aspen, jack pine and black spruce stands (Dang et al. 1997b, Sullivan et al. 1997), and values reported in the literature for representative cover types (Van Cleve et al. 1983, Aerts et al. 1992, Mitsch and Gosselink 1993, Schimel et al. 1996). Soil rooting depth and water holding capacity information were derived for each land cover class from a 1:10 6 scale digital soils inventory database of Canada (Acton et al. 1991), as well as volumetric soil water and soil survey measurements conducted at several sites within and adjacent to the SMSA (Boreas Science Team 1995, Cuenca et al. 1997). Soil b-parameter values define the slope of the functional PSI response to changes in soil water and were derived from values reported in the literature for representative soil types (Cosby et al. 1984). For this investigation, soil structural characteristics were defined for each vegetation type and assumed to be constant within the area represented by each land cover class. BIOME-BGC uses daily maximum and minimum air temperatures, solar irradiance and precipitation to determine daily carbon and water fluxes. Daily meteorological data were interpolated over a 1-km resolution digital elevation map (DEM) of the SMSA using a daily meteorological interpolator (Running et al. 1987, Thornton et al. 1997), digital elevation information (i.e., elevation, slope, aspect), and daily meteorological data from approximately 60 weather stations within the BOREAS region. Gridded daily meteorological data were produced for the 3-year ( ) study period. Meteorological data were obtained from the National Climatic Data Center s Global Surface Summary of the Day database, the Saskatchewan Research Council s Automatic Meteorological Stations database and BOREAS tower flux site measurements (National Weather Service 1988, BOREAS science team 1995, Shewchuk 1997). The DEM information were provided by the BOREAS HYD-08 Team and BOREAS staff (BOREAS Science Team 1995). Daily model simulations were conducted within the study region from January 1, 1994 to December 31, Meteorological conditions were defined from the 3-year gridded daily meteorological fields, whereas 1994 land cover type and biomass maps defined surface physical conditions for the 3-year period. Spatially distributed estimates of initial soil water and snow water equivalent depth were required to initialize the 1994 water balance. The model was initialized with a uniform, snow water equivalent depth of 3.3 cm and a soil water content set at 95% of field capacity. These values were determined from BIOME-BGC point simulations at BOREAS SSA tower sites and 1993 daily meteorological data from Nipawin Airport (~53 20 N, W) near the southeast corner of the SSA. BIOME-BGC-simulated dates of spring snow depletion were used in conjunction with 10-year records ( ) of estimated annual aboveground NPP to explore the relationship between NPP and interannual variability in the timing of spring thaw. The estimated date of snow cover depletion in the spring was used as an indicator of spring thaw timing, which defines the seasonal shift from frozen to non-frozen (i.e., the TREE PHYSIOLOGY ON-LINE at

6 766 KIMBALL, KEYSER, RUNNING AND SAATCHI presence of liquid water in soil, snow and vegetation) conditions. This period can occur approximately 1 5 weeks before the snow cover depletion date, depending on temperature and precipitation conditions. Annual aboveground NPP (ANPP) information was obtained for 57 sites within the BOREAS northern (NSA) and southern study regions. Detailed discussion of experimental design, sampling methods and analyses of these data is provided by Gower et al. (1997). These sites were selected to represent dominant black spruce, jack pine, aspen and mixed (broadleaf deciduous evergreen conifer) boreal forest cover types. Species-specific allometric relationships were developed from stem diameter, destructive biomass harvest and litterfall measurements collected during the 1994 growing season. Measurements were structured to characterize an approximately 30-m 2 area at each site location. The ANPP values were then estimated for the previous 10 years from annual stemwood radial increment measurements, and foliar and litter biomass allometric relationships (Gower et al. 1997). The ANPP results for dominant forest classes within each study region were then averaged for each year to obtain mean, biomelevel estimates for comparison with BIOME-BGC NSA and SSA snow cover simulations. Gower et al. (1997) provide a detailed discussion of potential error sources associated with ANPP estimates, but do not provide specific quantification of the uncertainty associated with these results, because this would require either the development of new allometric equations or installation of new plots for validation purposes, or both. Potential sources of error are primarily associated with the application of allometric relationships to derive ANPP for previous years and at sites where foliar and litter biomass data were not collected. Application of allometric relationships obtained from the literature for similar cover types, however, showed generally good agreement (within 10%) with total biomass at all sites (Gower and Vogel 1999). BIOME-BGC simulations of snow cover depletion dates were obtained using daily weather records from the Thompson (~55 48 N, W) and Nipawin airport atmospheric weather stations within the BOREAS SSA and NSA study regions (National Weather Service 1988). On average, depletion of spring snow cover occurs during the last week of April at the Nipawin station, but can range over a 7-week period (April 3 May 22) in any given year, according to long term ( ) daily snow depth records. Spring snow depletion at the Thompson station exhibits a similar range of interannual variation as the Nipawin station, but is generally delayed by approximately 1 2 weeks. Snow cover simulations derived using the Thompson weather station data were compared with NSA ANPP values, whereas model results derived from Nipawin weather station data were compared with SSA ANPP values. Snow cover simulations were previously compared with NSA and SSA tower site snow depth measurements for 1994, and were found to correspond reasonably well (Kimball et al. 1997a). Subsequent comparisons were also conducted using Nipawin and Thompson station 23-year daily snow depth records. Model results accounted for 76.9% of the daily variability in measured snow depths (P < 0.001) and were generally within 4 (± 3) days of observed spring snow depletion dates and 3 (± 2) days of fall snow appearance dates. Results and discussion Regional meteorological conditions Meteorological conditions from 1994 to 1996 were quite different relative to long-term (23-year) weather conditions for the region (Table 2). Year 1994 had relatively warm spring and fall conditions, with a cooler summer and more precipitation than normal. Years 1995 and 1996 had generally cooler spring and fall conditions, but higher summer temperatures than 1994 or the long-term record. Summer conditions for 1995, however, were drier than 1994, 1996 or the long-term record. Spatial variability in gridded air temperature and solar irradiance represented less than 6% (0.3 C) and 0.9% (1.5 W m 2 ) of mean annual results, respectively. The gentle topography of the region and similar daily meteorological conditions for weather stations in and around the study region were generally responsible for the low spatial heterogeneity in these results. Spatial variability in annual precipitation was also low, representing less than 7% (0.4 cm) of the mean annual total precipitation for the region. However, variability in daily precipitation was much greater because of the predominance of small, spatially erratic precipitation events produced by convective activity during the summer months. Spatial heterogeneity in daily meteorological conditions were likely much greater than our results indicate because of the influence of sub-grid scale microtopography, vegetation and soil water effects on the surface energy balance. These effects may influence regional carbon fluxes, but were not addressed within the framework of this investigation because of the coarse nature of the DEM and scarcity of surface weather stations (< 1 station per 1900 km 2 ). AIRSAR-defined land cover conditions Seven land cover classes were distinguished within the study region from the AIRSAR land cover map (Figure 2). Wet conifer (WC) was the dominant cover class, representing 45.9% of the study area, whereas DC covered 17.1% of the study region and represented the next most dominant cover class. Other classes constituted from 2.2% (OW) to 13.8% (WD) of the region and were generally more dispersed and fragmented than WC areas. The accuracy level of this classification was estimated to be greater than 90%, based on 1994 ground survey measurements across 19 forest stands and a digital vegetation map derived from infrared aerial photography (Saatchi and Rignot 1997). The AIRSAR-derived biomass and LAI distributions were quite variable within the SMSA, both within and between different land cover classes (Table 3). These variables were approximately normally distributed and significantly different (P < 0.001) between land cover classes. Within-class biomass TREE PHYSIOLOGY VOLUME 20, 2000

7 REGIONAL ASSESSMENT OF BOREAL NPP 767 Table 2. Weather summary for the Nipawin Airport weather station (~53 20 N, W) located near the SE corner of the BOREAS SSA; data represent mean annual weather conditions for , as well as long-term ( ) means and standard deviations (SD) for the site; air temperature data are also summarized for fall (September 1 November 30), spring (March 1 May 31) and summer (June 1 August 31) conditions. Characteristic SD Mean air temperature ( C) Mean air temperature ( C, fall) Mean air temperature ( C, spring) Mean air temperature ( C, summer) Total precipitation (cm) Total rainfall (cm) Total snow water equivalent (cm) diversity was large, however, with mean coefficients of variation ranging from 9% (MX) to 69% (DSTRB). Deciduous (DEC) stands generally had the most aboveground biomass even though WC stands tended to have more crown biomass and leaf area. Disturbed sites created as a result of fire and logging had the lowest biomass and leaf area of all the classes. Figure 2. The AIRSAR-derived landcover classification map of the study area at a 30-m spatial scale. The image is in the BOREAS grid system (BOREAS Science Team 1995), which is an Albers EqualArea Conic projection. The lower graph depicts the proportion of classified area represented by each landcover class within the AIRSAR map. The accuracy level of this classification was estimated to be greater than 90%, based on 1994 ground survey measurements across 19 forest stands and a digital vegetation map derived from infrared aerial photography (Saatchi and Rignot 1997). Biomass values within WD areas were similar to WC conditions and are generally in the upper range of biomass values reported for Siberian and Western European wetlands, but are similar to values reported for forested fens, bogs and peatlands in northern Minnesota and Michigan (Mitsch and Gosselink 1993). The AIRSAR results were generally consistent with the ranges of other values obtained from allometric relationships, biomass and optical LAI field measurements within the BOREAS region (e.g., Chen et al. 1997, Gower et al. 1997, Saatchi and Moghaddam 2000a, 2000b). For example, Gower et al. (1997) estimated LAI (defined as 50% of total LAI) values for 1994 on the order of 5.6 (± 1.7), 2.8 (± 0.8) and 3.3 (± 0.7) for mature black spruce, jack pine and aspen stands, respectively, within the SSA from biomass harvest plot information. Gower et al. (1997) also reported respective total aboveground biomass for these stands of approximately 49, 31 and 93 Mg C ha 1. These values were similar in relative magnitude and range to AIRSAR-derived LAI for WC, DC and DEC cover classes, respectively (Table 3). The AIRSAR results were also compared directly with biomass harvest plot measurements by extracting mean, maximum and minimum values from AIRSAR 3 3-pixel windows centered over approximately 13 tower, carbon evaluation and auxiliary harvest plot locations within the SMSA (Gower et al. 1997, Sellers et al. 1997). Biomass sampling at these sites was structured to characterize vegetation within an approximate m area from a network of 1 4 replicate plots ranging in size from 56 m 2 to 900 m 2, depending on tree densities (Gower et al. 1997). Scatterplots of AIRSAR and harvest plot results showed a large degree of within-site variability, with differences between maximum and minimum values averaging 53 and 75% of mean results for AIRSAR windows and harvest plots, respectively (Figure 3). The ranges of biomass and LAI values were similar between AIRSAR and harvest plot results, though absolute errors for individual sites were large, averaging 46 (14.5 Mg C ha 1), 110 (2.2 Mg C ha 1) and 34 (0.9 m 2 m 2) percent for stem biomass, foliar biomass and LAI, respectively. Differences between AIRSAR and harvest plot results at individual sites were attributed to the extreme spatial complexity of sur- TREE PHYSIOLOGY ON-LINE at

8 768 KIMBALL, KEYSER, RUNNING AND SAATCHI Table 3. Statistical summary (mean with 1 standard deviation in parenthesis) of SMSA crown, stem and foliar biomass 1 (MgCha 1 ), and leaf area index (LAI) for wetland (WD), mixed forest (MX), dry conifer (DC), wet conifer (WC), disturbed (DSTRB) and deciduous (DEC) landcover types, as well as the entire region within the SMSA. Crown and stem biomass values were obtained directly from AIRSAR remote sensing data, whereas LAI and foliar biomass were derived indirectly from biomass maps and allometric relationships obtained from biomass harvest plot measurements. Parameter WD MX DC WC DSTRB DEC Region Crown biomass 15 (3) 21 (2) 12 (2) 18 (3) 3 (2) 15 (2) 16 (4) Stem biomass 71 (16) 87 (8) 52 (17) 70 (20) 117 (17) 74 (26) LAI 3.3 (0.6) 4.5 (0.4) 2.5 (0.3) 4.3 (0.8) 0.6 (0.4) 3.3 (0.4) 3.7 (0.9) Foliar biomass 1.5 (0.3) 2.5 (0.2) 2.0 (0.3) 3.5 (0.6) 0.5 (0.2) 1.0 (0.1) 2.5 (1.0) 1 All values have been converted to Mg C ha 1 assuming that biomass is 50% carbon. face vegetation patterns within the study region. This complexity affected the characterization of vegetation distributions from both AIRSAR and destructive biomass harvest techniques, because non-site-specific allometric relationships such as mean SLA and foliar to crown biomass ratios were used to interpret AIRSAR results and spatially extrapolate field measurements within different vegetation communities. Figure 3. Scatterplots of AIRSAR versus biomass harvest plot stem biomass (Mg C ha 1 ) and LAI (defined as 50% of total LAI) estimates for The AIRSAR results were obtained by extracting mean, maximum and minimum values from 3 3-pixel windows centered over approximately 13 tower site, carbon evaluation and auxiliary harvest plot locations within the study area. Biomass sampling within each plot was structured to characterize vegetation within an approximate m area from a network of 1 4 replicate plots ranging in size from 56 m 2 to 900 m 2, depending on tree densities (Gower et al. 1997). Dotted lines depict the ranges (Max. Min.) of values reported within AIRSAR and harvest plot windows. NPP spatial complexity The NPP simulations were spatially complex, averaging 2.2 (± 0.56) Mg C ha 1 year 1 for 1994 (Figure 4). The results for each land cover class were approximately normally distributed and were significantly different among the six land cover types (P < 0.001). Within-class spatial variability was also large, with NPP coefficients of variation averaging 24% (± 0.34 Mg C ha 1 ) for the study period (Table 4). Net primary production was directly proportional to the amount of photosynthetic leaf area (Figure 5), which accounted for greater than 80% (P < 0.001) of the variation in annual NPP simulations. The direct, linear correspondence between NPP and LAI, and relative differences in the slopes of the NPP LAI relationships between conifer and deciduous classes, are generally consistent with stand-level measurements of these variables within other North American temperate and boreal forests (e.g., Bonan 1993, Goetz and Prince 1996). The strong correlation between LAI and NPP results from the sensitivity of both variables to similar environmental constraints on resource allocation (Fassnacht and Gower 1997, Landsberg and Gower 1997). The apparent saturation of this relationship (Figure 5) above an LAI of approximately 5 is also evident in boreal stand measurements (e.g., Goetz and Prince 1996) and is predominantly a function of the model's respective linear and hyperbolic responses of respiration and net photosynthesis to LAI (Waring and Running 1998). Spatial complexity of NPP was strongly influenced by the magnitudes of aboveground biomass and estimated root carbon pools, which were directly proportional to autotrophic respiration rates (R m + R g ). Land cover classes with greater proportions of deciduous vegetation were generally associated with larger gross primary production rates per unit LAI because of larger photosynthetic capacities (Dang et al. 1997b). However, deciduous vegetation also had greater autotrophic respiration rates per unit leaf area compared with evergreen conifer stands (Ryan et al. 1997). Gross primary production for DEC, WD and MX cover classes averaged approximately TREE PHYSIOLOGY VOLUME 20, 2000

9 REGIONAL ASSESSMENT OF BOREAL NPP Figure 4. BIOME-BGC simulated NPP (Mg C ha 1 year 1) within the model grid for The image is in the same Albers Equal-Area Conic projection as the AIRSAR landcover map. The size of the model grid was defined by the aerial extent of available AIRSAR remote sensing data. White areas represent unknown surface landcover conditions that were masked from model analysis. Regional NPP averaged 2.2 (± 0.56) Mg C ha 1 year 1 for The MX stands were generally the most productive areas in the SMSA, averaging 3.4 Mg C ha 1 year 1, whereas other landcover types averaged from 2.7 (WD) to 1.0 (DSTRB) Mg C ha 1 year Mg C ha 1 year 1, whereas Rm averaged 7.1 Mg C ha 1 year 1 for Gross primary production and Rm were generally lower for WC and DC areas, with means of 8.7 Mg C and 5.6 Mg C ha 1 year 1, respectively. Measurements of CO2 eddy flux within the BOREAS SSA in 1994 also show both higher respiration and carbon (C) uptake rates for deciduous stands relative to evergreen conifer stands (Black et al. 1996, Baldocchi et al. 1997, Jarvis et al. 1997). Model NPP simulations indicated that MX stands were generally the most productive areas in the SMSA, averaging 3.4 Mg C ha 1 year 1 for 1994 (Table 4). Other land cover types averaged from 2.7 (WD) to 1.0 (DSTRB) Mg C ha 1 year 1. The relative magnitudes of annual NPP were generally Table 4. Annual summary (aerial mean with 1 standard deviation in parenthesis) of BIOME-BGC NPP results for wetland (WD), mixed forest (MX), dry conifer (DC), wet conifer (WC), disturbed (DSTRB) and deciduous (DEC) landcover classes within the SMSA. Cover type WD MX DC WC DSTRB DEC Total NPP (Mg C ha 1) 769 Figure 5. Scatterplots of BIOME-BGC estimated NPP (Mg C ha 1 year 1) for 1994 versus AIRSAR-derived LAI (defined as 50% of total LAI). These results represent a random sample (n = 1171) of approximately 1% of the study area population and show relationships for wet conifer (WC; 䊏), dry conifer (DC; +), fen wetland (WD; 䉫), deciduous (DEC; ), mixed deciduous evergreen (MX; 䉭) and disturbed (DSTRB; 䉬) AIRSAR-defined vegetation classes. Spatial variability of NPP was strongly controlled by the amount of aboveground biomass, particularly photosynthetic leaf-area, as well as biophysical differences between deciduous and coniferous vegetation. controlled by the amounts of photosynthetic leaf area and proportions of deciduous vegetation represented within each class. Although DEC areas were composed entirely of deciduous vegetation, these areas tended to have less photosynthetic leaf area, higher autotrophic respiration rates, and lower NPP than MX, WD or WC areas. The DSTRB areas had the lowest photosynthetic leaf area and associated NPP of all the classes. A summary of NPP values obtained from other studies of various forest types within the BOREAS region and related North American boreal forest communities is presented in Table 5. Annual NPP simulations for WC and DC areas were comparable in magnitude and range to NPP values derived from biomass harvest plot information for BOREAS black spruce and jack pine forest types (Gower et al. 1997, Steele et al. 1997). Estimates of NPP derived from mature aspen stand field measurements were generally larger than DEC simulations, and were more consistent with MX values. Biomass and NPP data for fen and other wetland areas were not available within the SMSA. However, model results were similar to the relative magnitudes of NPP values reported for northern bog marshes, forested fen and other peatland sites within Canada and the northern USA; reported values for these systems ranged from 1.4 to 9.7 Mg C ha 1 year 1 (e.g., Reader 1978) NPP temporal variability 2.75 (0.39) 3.37 (0.26) 1.91 (0.27) 2.05 (0.29) 1.00 (0.46) 1.95 (0.54) 2.16 (0.56) 2.19 (0.34) 2.74 (0.25) 1.56 (0.23) 1.89 (0.28) 0.88 (0.39) 1.06 (0.51) 1.79 (0.53) 1.95 (0.32) 2.50 (0.25) 1.67 (0.24) 1.84 (0.29) 0.80 (0.35) 0.72 (0.50) 1.69 (0.55) Simulated annual NPP distributions for the study region are presented in Figure 6. The approximate bimodal shapes of these distributions primarily reflect component interactions among relatively low productivity DC, moderately productive WC, and higher productivity MX classes. Model simulations showed significant reductions in regional NPP distributions of approximately 17% (0.4 Mg C ha 1) and 22% (0.5 Mg C ha 1) TREE PHYSIOLOGY ON-LINE at

10 770 KIMBALL, KEYSER, RUNNING AND SAATCHI Table 5. Net primary production (NPP) rates 1 for North American boreal forests. Land cover type Location Latitude NPP Reference 2 Jack pine Saskatchewan, Canada N , 2 Jack pine Manitoba, Canada N , 2 Young jack pine Michigan, USA N [ ] 3 Young jack pine Minnesota, USA N [ ] 3 Young jack pine Wisconsin, USA N [ ] 3 Black spruce Saskatchewan, Canada N , 2 Black spruce Manitoba, Canada N , 2 Black spruce Alaska, USA N , 5, 6 Black spruce Alaska, USA N (0.09) 4 Black spruce Minnesota, USA N [ ] 7 Bog Manitoba, Canada N Bog forest Manitoba, Canada N Bog forest Alberta, Canada N Bog forest Minnesota, USA N Poor fen Alberta, Canada N Rich fen Michigan, USA 44 N Fen forest Minnesota, USA 45 N Northern bog marsh North America 45 N [ ] 8 Aspen Saskatchewan, Canada N , 2 Aspen Manitoba, Canada N , 2 Aspen Minnesota, Canada N [ ] 7 Aspen Alaska, USA N (1.0) 4 1 All values have been converted to Mg C ha 1 year 1, assuming that wood, foliage and roots are 50% carbon; square brackets denote ranges of cited NPP values and round brackets denote standard deviations of cited NPP values. 2 References: 1, Gower et al. 1997; 2, Steele et al. 1997; 3, Zavitkovski et al. 1981; 4, Van Cleve et al. 1983; 5, Oechel and Van Cleve 1986; 6, Ruess et al. 1996; 7, Goetz and Prince 1996; 8, Reader 1978; 9, Reader and Stewart 1972; 10, Reiners 1972; 11, Richardson et al. 1976; 12, Grigal et al. 1985; 13, Szumigalski and Bayley Aboveground NPP. Figure 6. Simulated annual NPP distributions within the model grid for 1994, 1995 and Model results showed significant reductions in regional NPP distributions of approximately 17% (0.4 Mg C ha 1 ) and 22% (0.5 Mg C ha 1 ) for 1995 and 1996 relative to Stands with greater proportions of deciduous vegetation showed generally larger reductions in annual NPP than evergreen conifer stands. for 1995 and 1996, relative to Stands with greater proportions of deciduous vegetation showed larger reductions in annual NPP. The WD and MX stands, for example, showed reductions of approximately 23% (0.7 Mg C ha 1 ), whereas NPP for DEC stands was reduced by approximately 54% (1.1 Mg C ha 1 ). The DC and WC stands showed much smaller reductions averaging approximately 12% (0.2 Mg C ha 1 ). Reductions in NPP were predominantly caused by cooler spring conditions relative to 1994, which resulted in delayed leaf-out of deciduous vegetation and lower regional productivity during spring (Figure 7). In 1994, estimated leaf-out for deciduous vegetation was initiated at the end of April and completed during the third week of June. This pattern is consistent with 1994 observations within a mature SSA aspen stand (Black et al. 1996). In 1995 and 1996, spring air temperatures were 4 5 C lower (Table 2), and simulated snow cover remained on the ground approximately 1 month longer than in Snow depth measurements at the Nipawin atmospheric weather station also showed approximate 3- to 5-week delays in spring snow depletion for 1995 and 1996, relative to Estimated leaf-out for these years did not occur until the last week of May and the first week of June, whereas deciduous canopies were not in full leaf until mid-july. Canopy observation data within the SSA were not available for However, observations within a mature SSA aspen stand for 1996 also showed an apparent 5-week delay in the initiation of TREE PHYSIOLOGY VOLUME 20, 2000